Method for detecting the quantization of spectra

Abstract

The invention comprises the retrieval of a quantized spectrum of audio-data, which had been compressed conforming to psychoacoustic principles. This spectrum can be identified with the aid of sorted (and logarithmized) differences of the magnitude of the amplitude. A problem to be solved by the invention is to find the basis of sensitive indicators for the degree of quantization of a spectrum. The result of this calculation shows for a quantized spectrum several small values at amplitudes of the same quantization level, and obvious spikes at positions, where the amplitudes have reached a higher level of quantization, while a non-quantized spectrum typically resembles noise.

Description

BACKGROUND

Audio-data are being transmitted in a compressed way conforming to psychoacoustic principles and decoded by the receiver. Re-compressing of these decoded data conforming to psychoacoustic principles will lead to loss of quality. This loss can be omitted, if the parameters of the original compression can be detected and re-used, see e.g. EP-A-0746913.

A method for the detection of these parameters for codecs which use spectra, is testwise calculation of spectra with different parameters. If the parameters of analysis (start of transformation, window type, length of spectrum) match the codec parameters, the spectrum has a quantized appearance. Because of calculation impreciseness or different calculation precision of the codec and of the analysis calculation, the quantized spectrum will look slightly noisy, i.e. the amplitudes of the same level of quantization will NOT have the same magnitude.

The difficulty is to distinguish, whether a spectrum has a “quantized character”, so that different parameters of the spectrum can be analyzed concerning their impact on the quantization of the spectrum.

Because compression goes along with a vanishing of certain lines of the spectrum (cf. in Herre, Jürgen; Michael Schug: Analysis of Decompressed Audio—The “Inverse Decoder”, 109th AES Convention, Los Angeles 2000) those spectral lines are being counted, which are smaller than a threshold value.

INVENTION

The invention comprises the retrieval of a quantized spectrum of audio-data, which had been compressed conforming to psychoacoustic principles. This spectrum can be identified with the aid of sorted (and logarithmized) differences of the magnitude of the amplitude.

A problem to be solved by the invention is to find the basis of sensitive indicators for the degree of quantization of a spectrum. The result for a quantized spectrum shows several small values at amplitudes of the same quantization level, and obvious spikes at positions, where the amplitudes have reached a higher level of quantization, while a non-quantized spectrum typically resembles noise.

This problem is solved by the method disclosed in claim 1.

In principle, the inventive method is suited for tandem codecs (decoding-encoding). The invention can also be used at the procedure of reverse engineering with unknown codecs.

Advantageous additional embodiments of the invention are disclosed in the respective dependent claims.

DRAWINGS

Exemplary embodiments of the invention are described with reference to the accompanying drawings, which show in:

FIG. 1 Magnitudes of a quantized spectrum of a decoded audio signal;

FIG. 2 Magnitudes of a non-quantized spectrum;

FIG. 3 Amplitudes of FIG. 1) sorted;

FIG. 4 Amplitudes of FIG. 2) sorted;

FIG. 5 Differentiated sorted spectrum from FIG. 3);

FIG. 6 Differentiated sorted spectrum from FIG. 4);

FIG. 7 Simple block diagram of the detector.

EXEMPLARY EMBODIMENTS

FIG. 1 shows the magnitudes of a quantized spectrum of a decoded signal. The parameters of analysis (start of the transformation, window type, length of spectrum) match the codec parameters, therefore the spectrum has a quantized appearance.

In comparison to FIG. 1, FIG. 2 shows the magnitudes of a non-quantized spectrum. Thus, the parameters do not match the codec parameters.

FIG. 3 shows the sorted amplitudes from FIG. 1, while FIG. 4 shows the sorted amplitudes of a non-quantized spectrum from FIG. 2.

FIG. 5 shows the differentiated sorted spectrum from FIG. 3, while FIG. 6 shows the differentiated sorted spectrum of a non-quantized spectrum from FIG. 4

The comparison of FIGS. 5 and 6 will clearly show the difference between a quantized spectrum (FIG. 5) and a non-quantized spectrum (FIG. 6).

The next step is to calculate a scalar out of these spectra, which will be used as an indicator for the quantization.

The invention can be used in connection with an inverse encoder. Often an encoded and decoded audio signal has to be encoded for another transmission a second time. If the encoder parameters in the second encoder are the same as in the first, codec artefacts can be minimized. The inverse encoder tries to retrieve these parameters. In this case, for it has to be decided, if a spectrum looks quantized, which can be done with the help of the invention.

Another use of the inverse encoder and consequently also of the invention is in reengineering codecs for proving infringements in the area of intellectual properties.

FIG. 7 shows a simple block diagram of the invention. The purpose of the invention is to convert time domain samples in frequency domain, then sorting the amplitudes in frequency domain, followed by an optional logarithmizing of these sorted amplitudes and finally assessment of sorted spectra values, e.g. differentiation and detection of first peak.

Claims

1. Method of detecting the quantisation of spectra of a decoded audio signal in order to detect or retrieve encoding parameters that were used in the encoding of said audio signal, said method including the steps:

converting samples of the decoded audio signal into frequency domain, thereby using different test parameters like start position of transformation, window type, length of spectrum;

in the frequency domain, sorting the magnitudes of the spectral values;

assessing for a given set of test parameters whether the sorted spectral values have a quantised appearance.

2. Method according to claim 1, wherein the spectral values are logarithmised before said assessing step.

3. Method according to claim 1, wherein the sorted magnitudes of the spectra values are differentiated.

4. Method according to claim 3, wherein a sensitive indicators for the degree of quantisation of a spectrum is detected and said indicator is a function of the position and/or height of the first peak in said differentiation.

5. Method according to claim 4, wherein said indicator is the sum of the multiplication of the sorted spectra and the sorted spectra shifted by one line.

6. Apparatus for detecting the quantisation of spectra of a decoded audio signal in order to detect or retrieve encoding parameters that were used in the encoding of said audio signal, said apparatus including:

means for converting samples of the decoded audio signal into frequency domain, thereby using different test parameters like start position of transformation, window type, length of spectrum;

means for sorting, in the frequency domain, the magnitudes of the spectral values;

means for assessing for a given set of test parameters whether the sorted spectral values have a quantised appearance.